Methodology Development For Retrieving Land Surface Temperature And Near Suface Air Temperature Based On Thermal Infrared Remote Sensing

Both land surface temperature(LST)and near surface air temperature(NSAT)are key parameters in the physics of Earth surface,atmospheric system and surfaceatmosphere interaction processes,which play significant role in a variety of fields including meteorology,hydrology,and global change studies.Temperature information with fine spatial and temporal resolution is important for many remote sensing applications,such as drought forecasts,crop water stress estimation,plant physiology monitoring,crop yield estimation,numerical weather forecast,etc..Land surface temperature was listed as one of satellite measured parameters with high priority by the International Geo-Biosphere Programme(IGBP).Thermal infrared(TIR)Remote Sensing is an important mean to obtain thermal state of land surface and atmosphere at regional and global scales.Studies on thermal infrared Remote Sensing has made considerable development in last decade.At present,there are a dozen of TIR sensors in orbit,of which the highest spatial and temporary resolution reach 100 m and 30 minutes,respectively.With the extensive applications of TIR remote sensing,there are growing requirements in accuracy and spatial resolution of temperature information.Based on Landsat series satellite data at resolution between 60 m to 120 m and MODIS data at a resolution of 1000 m,this thesis focused on single channel LST retrieval,split window LST retrieval and NSAT mapping.The research contents and conclusions were as follows:1)The single channel algorithm corrects the effects of atmosphere to thermal infrared radiance emitted from land surface using the atmospheric parameters.Building accurate atmospheric parameters model is the key of single channel algorithm.An enhanced single-channel algorithm(SCen)was proposed for retrieving LST from Landsat series data(Landsat 4 to Landsat 8).The SCen algorithm includes three atmospheric functions(AFs),and the latitude and acquisition month of Landsat image were added to the AF models to improve the accuracy of AFs and then improve LST retrieval accuracy.Performance of the SCen algorithm was assessedwith simulated data,SURFRAD data and in situ measured LST,and accuracy of three single-channel algorithms(including the mono-window algorithm developed by Qin et al.,SCQin,and the generalized single-channel algorithm developed by Jiménez-Mu?oz and Sobrino,SCJ&S)were compared.The accuracy assessments with simulated data show root-mean-square deviations(RMSEs)for the SCen,SCJ&S,and SCQin algorithms were 1.36 K,1.86 K,and 2.51 K,respectively.Validation with SURFRAD data show RMSEs for the SCen and SCJ&S algorithms were 1.04 K and 1.49 K,respectively.Validation based on in situ measured LST show accuracy of the SCen and SCJ&S algorithms were 1.02 K and 1.47 K,respectively.2)Atmospheric water vapor content(AWVC)is a key input parameter of single channel algorithm for LST retrieval.For Landsat 4\5\7 with only one thermal infrared channel,AWVC can not be derived from satellite imagery itself.Landsat 8,newly launched Landsat satellite,has two thermal infrared channels.A NDVI based Split Window Covariance-Variance Ratio(SWCVR)algorithm was proposed for AWVC retrieval from Landsat 8 TIRS data,which has a better accuracy of about 0.18 g cm-2 compared to previous SWCVR.3)Simplification of the Planck's function,as a critical step in deriving the SWA,allows us to directly relate the radiance to the temperature toward resolving the radiative transfer equation(RTE)set.Instead of simplification of the Planck's function,in this thesis,Planck's radiance relationship between two adjacent thermal infrared(TIR)channels was modeled to resolve the RTE set.In addition,an atmosphere parameter representing the ratio of downward atmospheric radiance to upward atmospheric radiance was defined and introduced into the simplified RTE.A radiance based split window algorithm(RBSWA)was developed and applied to MODIS data.Performance of the proposed algorithm was assessed by using SURFRAD data,MODIS LST product(MOD11)and the simulated data produced by MODTRAN 4.0 with the TIGR3 database and the MODIS/UCSB emissivity library as inputs.Validation with simulated data show the RBSWA has an accuracy of 0.505 K,and achieved an improvement of 0.5 K compared to two famous SWAs.The sensitivity of RBSWA to input parameters was analyzed.The comprehensive LST error is about 1.05 K when the uncertainties are 0.02 Wm-2sr-1?m-1 for the atsensor radiance at band 31,0.006 for the land surface emissivity of band 31,0.6 g cm-2 for atmospheric water vapor content,6° for viewing zenith angle,and 0.5 K for RBSWA model.Cross validation with 108 MODIS LST imageries show a mean RMSE of 1.33 K.4)Near surface air temperature has high level spatial autocorrelation,as well as closely correlates to LST,the normalized difference vegetation index(NDVI),altitude and latitude etc..In this thesis,the geographically weighted regression(GWR)was introduced into NSAT mapping.GWR utilizes both NSAT's autocorrelation and its correlation to other Remote Sensing and Geographical Information Systems variables.Compared to the standard multiple linear regression,the accuracy of NSAT mapping based on GWR achieved an obvious improvement of 4 ?.The accuracy of NSAT mapping based on GWR is better than Kriging interpolating model in warm months;while the performance of NSAT mapping based on GWR is lower than Kriging interpolating model in cold months.In warm months,such as June,July,August and September in the northern hemisphere,the monthly minimum,average and maximum NSAT mapping based on GWR have improvements of 0.3,0.4 and 0.5 ?,respectively.In conclusion,the accuracy of method based on GWR is slightly better than Kriging for whole year;and obviously better than Kriging interpolating model for warm month.In order to obtain higher accuracy of NAST mapping,the optimal method need to be selected based on different levels of NSAT.